Yoojin-nam/MeducAI: v1.1 — Paper 1 Submission Archive + Evaluation Dataset
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Cumulative archive for Paper 1 ("Sequential multi-stage validation of large language model–generated radiology learning content," in revision at npj Digital Medicine). v1.0 (2026-03-30, retained): Protocol freeze, prompt versions S5R0–S5R3, analysis scripts. Original archive of the GitHub release tag v1.0.0-protocol-freeze. v1.1 (this version): Adds the full peer-review-time evaluation dataset (CC BY 4.0): cards_analysis.csv — 1,284 evaluated cards with S5 outputs (decisions, technical accuracy, educational quality, image-quality sub-scores, regen status). ratings_deidentified.csv — de-identified human evaluator ratings (rater × card cells). rater_summary.csv, export_manifest__regen.json — auxiliary metadata. images/final_cards/ — 1,284 baseline (pre-regen, optimized-for-Anki) card images. images/regen_after/ — 323 raw IMAGE_REGEN / CARD_REGEN outputs. Deployment mapping (which image was actually shown to learners). The pipeline writes baseline and regenerated images to separate folders; the Anki export script (3_Code/src/tools/anki/export_final_anki_integrated.py) selects per card based on the S5 outcome: PASS (961 cards) → deployed image = images/final_cards/.jpg. IMAGE_REGEN (62 cards) and CARD_REGEN (261 cards) → deployed image = images/regen_after/_regen.jpg (the regenerated variant; the baseline at final_cards/ for these 323 cards was rejected by S5 and not deployed). The cards_analysis.csv::image_filename column always references the baseline filename. For regenerated cards, the actually-deployed image is at images/regen_after/ with the same basename plus the _regen.jpg suffix; this is implied by s5_was_regenerated == 1 (323 rows). Manuscript figure rendering. Main Figure 1 panels c, d and Supplementary Figure S2 panels a–f are rendered from the raw, pre-optimization S4 / IMAGE_REGEN outputs (high-resolution RGB / grayscale). Visual content matches the optimized images in this archive; the byte-level files differ only in resolution and colour encoding (see 3_Code/src/tools/optimize_images.py on GitHub). See README.md for full schema, license, and reproducibility instructions. The companion code repository is at github.com/Yoojin-nam/MeducAI (Apache-2.0).
Fields of Study & Subjects
Artificial Intelligence in Healthcare and Education
Innovations in Medical Education
Artificial Intelligence in Education
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 30 March 2026
- Language
- EN